Semantic modeling enables you to intelligently manage and query semantic models, metrics, and dimensions using natural language. Buddy automatically understands business intent, creates or modifies semantic objects, and supports querying semantic metadata in natural language (including similarity detection), helping analysts quickly understand and use the metric system.
Overview
Semantic modeling is a core infrastructure for data analysis. Before conducting data analysis, business analysts typically need to create metrics, manage existing semantic models and dimensions, or quickly locate required information when faced with a large number of existing metrics. Traditional methods require repeated operations across multiple pages, resulting in a steep learning curve. This feature covers the following scenarios:
Semantic Model Management: Create, modify, and query semantic models using natural language, including configurations such as table relationships, metric definitions, and dimension bindings.
Metric Management: Create atomic metrics, derived metrics, and composite metrics using natural language, configure attributes such as metric definitions, calculation logic, and time granularity, and support batch creation and modification.
Dimension Management: Create and manage dimensions, dimension attributes, and hierarchical relationships using natural language, and configure associations between dimensions and fact tables.
Natural Language Query of Semantic Metadata: Describe your requirements in natural language, and AI performs semantic search across semantic layer metadata to return matching models, metrics, and dimension information. It also supports similarity detection to identify metrics with similar names or definitions, preventing duplicate definitions.
Metric System Exploration: New analysts or users unfamiliar with the metric system can learn about the existing metric system structure, relationships between metrics, and usage methods through conversational interactions.
Prerequisites
Buddy has been enabled for the workspace.
The current account has Buddy development mode permissions.
You have view permissions for the semantic models, metrics, and dimensions that need to be managed.
To create or modify semantic objects, you must have edit permissions for the semantic layer.
The semantic layer already contains basic data models (physical tables or views) that can be used as underlying tables when semantic models are created.
Information You Must Provide
Before performing semantic modeling operations, Buddy will ask you to provide the following minimum information set, while the rest is automatically recommended by AI:
|
Operation target | Object types to be managed (semantic models / metrics / dimensions) and specific names |
Operation intent | Create / Modify / Query / Explore (can be implicitly expressed through natural language) |
If your input is incomplete, Buddy will ask for the missing information through an "Ask User card" before continuing. If your input is complete, Buddy will execute directly and display the results.
Configurations That AI Can Intelligently Generate
|
Metric definition | Automatically generate SQL calculation logic and aggregation methods based on the user's natural language description. |
Metric Attribute | Recommend data types, time granularities, and statistical periods based on the metric name and caliber description. |
Dimension Association | Analyze the fields involved in the metric calculation logic and automatically recommend associated dimensions. |
Model Association | Recommend relationships between fact tables and dimension tables based on foreign keys, naming conventions, and field types. |
Similar Metric Detection | Perform similarity matching on the name and caliber of the new metric, and prompt for existing similar metrics. |
Metric Classification Suggestion | Recommend the metric category based on the metric name and business semantics. |
Operation Steps
Step 1: Expressing Semantic Modeling Requirements
Describe your requirements in natural language in the Buddy input box, and Buddy will automatically identify the operation type and target object.
Example Input:
Metric Management Scenario:
Create a "Daily Active Users" metric for me based on the distinct count of user_id in the dwd_user_login table.
To modify the calculation definition of the "GMV" metric, refund orders must be excluded.
Configure two time granularities, daily and weekly, for the "Order Conversion Rate" metric.
Semantic Model Management Scenario:
Create an order analysis semantic model based on dwd_order_detail and dim_product.
Add an "Average Order Value" measure to the order analysis model.
Dimension Management Scenario:
Create a "Region" dimension that includes three levels: province, city, and district.
Associate the "Product Category" dimension with the order analysis model.
Semantic Metadata Query Scenario:
Are there any metrics related to "User Activity"?
Find all metrics whose calculation definition includes "Payment Amount".
Are "DAU" and "Daily Active Users" the same metric?
Metric System Exploration Scenario:
How is our metric system organized? What are the major categories?
What are the core metrics in the transaction domain? What are the relationships between them?
What is the definition of the "GMV" metric? What are its derived metrics?
Step 2: Completing Missing Information
If the input is incomplete, Buddy will ask you for the missing information through an Ask User card:
If you create a metric without specifying a base table, the available fact tables are listed for you to choose from.
If a metric is created but its definition is not clear enough, you are required to provide the calculation logic.
If you create a dimension without specifying an associated table or primary key, you will be asked to confirm.
After you answer, Buddy will continue the execution.
Step 3: Similarity Detection (Automatically Triggered During Scenario Creation)
When you request to create a new metric or dimension, Buddy automatically performs a similarity check:
Name Similarity: Checks whether existing metrics and dimensions contain objects with similar names, such as "Daily Active Users" and "DAU".
Definition Similarity: Compares whether the calculation logic is highly similar to existing metrics to avoid duplicate definitions.
Detection Result Display:
If no similar objects are found, proceed directly to the next step.
If similar objects are found, a list of similar items is displayed:
Similar metric and dimension names
Similarity Score
Definition Comparison
Buddy suggests: Reuse existing ones or confirm as new.
Note:
Similarity detection helps prevent duplicate metric definitions and maintain consistency and standardization across the metric system.
Step 4: Reviewing AI-Generated Configurations
Buddy displays the configurations of semantic objects to be created or modified in structured cards:
Metric Creation:
Metric name and metric code
Metric type (atomic, derived, or composite)
Calculation Definition (SQL Expression)
Aggregation method and data type
Time granularity and statistical period
Associated Dimensions
Associated Catalog
Semantic Model Creation:
Model name and associated base table
Join relationships between fact tables and dimension tables (join conditions)
Measure list and dimension list
Dimension Creation:
Dimension name and dimension code
Dimension attribute list
Hierarchy (for example, province → city → district)
Associated table and primary key
Note:
You can modify any configuration on the card before confirming, and Buddy will apply the modified configuration.
Step 5: Confirming and Applying Configurations
After you confirm the configuration, Buddy calls the DataBuddy semantic layer API to perform create and modify operations and displays:
Operation result (success or failure)
ID and status of the created object
Failed items and reasons (such as naming conflicts, insufficient permissions, or missing base tables)
Step 6: Linking Subsequent Operations (Optional)
After completing semantic modeling, Buddy proactively suggests next steps based on the scenario:
|
Publish Metric | The metric has been created. Do you want to publish it to the production environment? |
Associate Quality Rule | |
Continue Exploring | Do you want to view the derived metrics or upstream and downstream relationships of this metric? |
Key Milestone Confirmation
|
Object | Before operation | Confirm the semantic object to be created or modified. |
Similarity Detection Result | When creating a metric | Review similar metrics and determine whether to reuse or create new ones. |
Configuration Details | Before creation or modification | Review metric calibers, dimension associations, and model configurations. |
Publish Operation | Before publishing | Confirm whether to publish the semantic object to the production environment. |
FAQs
Q: What Types of Metrics Does Buddy Support Creating?
A: Supports creating atomic metrics, derived metrics, and composite metrics. Atomic metrics define basic measurement calculations (such as COUNT(DISTINCT user_id)). Derived metrics add time restrictions and dimension filters to atomic metrics (such as "daily active users in the last 7 days"). Composite metrics are calculated by combining multiple metrics (such as "conversion rate = paying users / visiting users").
Q: What Are the Criteria for Similarity Detection?
A: Buddy performs detection from two dimensions: name similarity and definition similarity. Name similarity is based on semantic matching, for example, "DAU" and "Daily Active Users" are determined as highly similar. Definition similarity is based on equivalence analysis of SQL logic. When the similarity of either dimension exceeds the threshold, a prompt is triggered.
Q: Is There a Difference Between Metrics Created via Buddy and Those Created Manually on the Semantic Layer Page?
A: There is no difference at all. Buddy creates metrics through the DataBuddy semantic layer API, and the created metrics are in the same management system as manually created ones. You can view and manage them centrally on the semantic layer page.
Q: How Can New Analysts Quickly Understand the Metrics System?
A: You can ask Buddy questions directly in natural language, such as "How is our metrics system organized?", "What are the core metrics in the transaction domain?", and "What is the calculation definition of GMV?". Buddy will introduce the metrics system structure, metric meanings, and relationships between metrics in a conversational manner, helping new users get started quickly.
Q: What Is the Difference Between Semantic Metadata Query and Search in Data Catalog?
A: Data catalog search targets the physical layer (tables and fields), while semantic metadata query targets the semantic layer (models, metrics, and dimensions). Buddy's semantic query performs matching based on semantic understanding, can understand equivalent expressions of business meanings (for example, searching for "user activity" can match the "DAU" metric), and supports similarity comparison at the definition level.
Q: Is There a Limit on the Number of Metrics Created in Batches?
A: In a single conversation, Buddy recommends creating no more than 20 metrics at a time to ensure that the definition and configuration of each metric can be thoroughly reviewed. If you need to create a large number of metrics, it is recommended to complete the creation across multiple conversations.
References